The use of a multichannel head for the high-speed formation of latent images inevitably results in poorer picture quality due to insufficient machining precision of the head or insufficient precision of the head-moving mechanism. That is, it is difficult to provide channels having an equal tracking width. It is also difficult to eliminate errors that stem from the tracking differences of the channels and the moving quantity of subscanning. The latter problem involves not only errors resulting from machining precision but also vibration resulting when the recording head moves in the subscanning direction. Therefore, subscanning involves variance in the feeding pitch.